91Ó°ÊÓ

In hypothesis testing, especially when comparing two proportions from different groups, we often use a concept called the **pooled proportion**. This is a method to find a single estimated proportion that represents both groups combined. It's like taking all the data together and figuring out what proportion is more representative for the entire dataset rather than individual observations.
To calculate it, we combine successes from both groups and divide by the total number of observations from both groups. Here's how it's done:

• Add up the 'successes' from both groups, which are the numerator values in the proportions.
• Add up all the observations from both groups.
• Divide the combined 'successes' by the combined total number of observations.

Let's see this in action with our data: The successes (deaths) were 347 for the placebo and 327 for aspirin. Adding these gives us 674. The total number of observations (people) for placebo was 11,535 and for aspirin, it was 14,035. Adding these gives us a total of 25,570. The pooled proportion is then calculated by dividing the total successes by the total population, yielding 0.026. This proportion is used to assess if there is a significant difference between the groups under the null hypothesis.

The **standard error** is a measure that tells us how much the sample proportion is expected to vary from the true population proportion. It's like a margin of error that helps us determine the reliability of our estimate. Typically, a smaller standard error indicates more reliable estimates.
To calculate the standard error for the difference between two proportions, use the formula:
\[ SE = \sqrt{\hat{p}(1 - \hat{p}) \left( \frac{1}{n_1} + \frac{1}{n_2} \right)} \]
Where:

• \( \hat{p} \) is our pooled proportion.
• \( n_1 \) and \( n_2 \) are the sample sizes for the placebo and aspirin groups respectively.

In our exercise:

• \( \hat{p} = 0.026 \)
• \( n_1 = 11,535 \) and \( n_2 = 14,035 \)

Plugging these into the formula gives us the standard error of 0.002. This small standard error highlights that our estimated proportion differences are quite precise and not likely to stray far from the true values.

The **Z-test statistic** helps us determine how far our observed results are from the null hypothesis, in terms of standard deviation units. It's a way to quantify the difference and see if it's big enough to be considered unlikely just by chance.
To calculate the Z-test statistic for two proportions, use this formula:
\[ z = \frac{\hat{p}_1 - \hat{p}_2}{SE} \]
Where:

• \( \hat{p}_1 \) is the sample proportion for the first group (placebo).
• \( \hat{p}_2 \) is the sample proportion for the second group (aspirin).
• \( SE \) is the standard error of the difference between proportions that we have calculated earlier.

For our specific example, we took \( \hat{p}_1 = 0.0301 \) and \( \hat{p}_2 = 0.0233 \), with the standard error being 0.002.Substituting these into our formula, we get a test statistic value \( z = 3.4 \). This means the observed difference is 3.4 standard deviations away from zero (or no difference). A value like 3.4 is quite large, suggesting that the observed difference is not easily attributed to random chance.

When conducting hypothesis testing, we use the **P-value** to determine the strength of the evidence against the null hypothesis. It's essentially the probability of observing data as extreme as what we have, assuming the null hypothesis is true. Lower P-values indicate stronger evidence against the null hypothesis.
In the context of our exercise, the P-value is extracted from the Z-test statistic. A high Z-value, like 3.4, translates to a low P-value, usually below most common significance thresholds like 0.05 or 0.01.
This implies:

• The chance of observing such a difference in proportions, assuming no real difference (under the null hypothesis), is very low.
• This provides strong evidence to reject the null hypothesis.

In our cancer death rate study, the low P-value suggests that the difference in death rates between the placebo and aspirin groups is statistically significant. In practical terms, this means there is likely a real difference in death rates related to the treatment method, not merely due to sampling variability.